BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Exploring the Linear Subspace Hypothesis in Gender Bias Mitigation
  - Francisco Vargas Palomo
DTSTART:20201207T163000Z
DTEND:20201207T170000Z
UID:TALK154246@talks.cam.ac.uk
CONTACT:Bingqing Cheng
DESCRIPTION:Paper: https://www.aclweb.org/anthology/2020.emnlp-main.232.pd
 f\n\nAbstract: Bolukbasi et al. (2016) presents one of the first gender bi
 as mitigation techniques for word embeddings. Their method takes pre-train
 ed word embeddings as input and attempts to isolate a linear subspace that
  captures most of the gender bias in the embeddings. As judged by an analo
 gical evaluation task\, their method virtually eliminates gender bias in t
 he embeddings. However\, an implicit and untested assumption of their meth
 od is that the bias subspace is actually linear. In this work\, we general
 ize their method to a kernelized\, non-linear version. We take inspiration
  from kernel principal component analysis and derive a non-linear bias iso
 lation technique. We discuss and overcome some of the practical drawbacks 
 of our method for non-linear gender bias mitigation in word embeddings and
  analyze empirically whether the bias subspace is actually linear. Our ana
 lysis shows that gender bias is in fact well captured by a linear subspace
 \, justifying the assumption of Bolukbasi et al. (2016).
LOCATION:virtual ZOOM meeting ID: 263 591 6003\, Passcode: 000042\, https:
 //us02web.zoom.us/j/2635916003?pwd=ZlBEQnRENGwxNmJGMENGMWxjak5nUT09
END:VEVENT
END:VCALENDAR